WO2020248345A1 - 一种基于分治思想的疵病宽度实时溯源方法 - Google Patents
一种基于分治思想的疵病宽度实时溯源方法 Download PDFInfo
- Publication number
- WO2020248345A1 WO2020248345A1 PCT/CN2019/099008 CN2019099008W WO2020248345A1 WO 2020248345 A1 WO2020248345 A1 WO 2020248345A1 CN 2019099008 W CN2019099008 W CN 2019099008W WO 2020248345 A1 WO2020248345 A1 WO 2020248345A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- defect
- width
- imaging
- distribution
- dark field
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/0006—Industrial image inspection using a design-rule based approach
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10056—Microscopic image
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20004—Adaptive image processing
- G06T2207/20012—Locally adaptive
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30164—Workpiece; Machine component
Definitions
- the invention belongs to the technical field of machine vision detection, and specifically relates to a method for real-time tracing of defect width based on a divide-and-conquer idea.
- the calibration transfer function is used to trace the width of the defect in real time; when the width of the defect is close to the diffraction limit of the system, the calibration transfer function is difficult to measure the width of the defect
- the problem of accurate source tracing is to use the multi-dimensional feature vector in the electromagnetic field simulation database of the blemish scattering dark field imaging as the data basis, and train the extreme random tree regression model to trace the width of the flaw close to the diffraction limit; for the smaller flaw, use it Adaptive threshold segmentation algorithm for traceability. Combining the above three situations, a real-time traceability method of defect width based on divide and conquer idea with excellent portability is proposed to objectively detect
- the purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for real-time traceability of the defect width applied to the microscopic scattering dark-field imaging detection environment to improve the recognition accuracy of the defect width.
- Step 1 Estimate the defect width according to the calibration transfer function of the standard defect digital calibration technology, and record the CTF width as w CTF . If w CTF is greater than 5 ⁇ m, then the defect width conforms to the geometric imaging law and is located in the linear segment of the calibration transfer function. It is more accurate to use w CTF as the estimated value of the actual defect width; if w CTF is less than or equal to 5 ⁇ m, then the defect If the width of the disease is small, the diffraction broadening phenomenon will occur during imaging, which no longer conforms to the geometric imaging law. The error of using w CTF as the actual width of the defect is large, and then skip to step 2.
- Step 2 According to the multi-dimensional feature vector analysis technology based on the electromagnetic field simulation database of the scattered dark field imaging of the defect, estimate the width of the defect and record it as the regression width w regrs . If the regression width w regrs is greater than the system diffraction limit ⁇ 0 , it is more accurate to use the defect regression width w regrs as the estimated value of the actual defect width; if the regression width w regrs is less than or equal to the system diffraction limit ⁇ 0 , then the defect regression width w Regrs cannot distinguish the different width values of this width range, and step 3 is needed for further identification.
- Step 3 Process the dark field image according to the adaptive threshold segmentation method, extract the defect width, and record it as the imaging width w img . If the imaging width w img is greater than the system diffraction limit ⁇ 0 , it is regarded as a singular point, and the regression width w regrs is still used as the estimated value of the actual width; if the imaging width w img is less than or equal to the system diffraction limit ⁇ 0 , the defect imaging width is used w img as an estimate of the actual width.
- the defect width w CTF is obtained by establishing a calibration transfer function.
- the calibration transfer function describes the relationship between the actual width of the defect and the pixel width after the defect is imaged.
- the acquisition process is as follows:
- the pixel width of the defect has a linear relationship with the actual width. Therefore, the width traceability of most defects can be achieved by calibrating the transfer function.
- the geometrical optical imaging theory is no longer applicable at this time. Fine defects will cause diffraction effects, and the pixel width will be widened. This will cause the same pixel width to correspond to more than one actual width when using the calibrated transfer function for width traceability, which will cause confusion and the actual width cannot be accurately obtained.
- the calibration transfer function no longer satisfies the linear relationship, and the width traceability is confused. The width traceability of this range needs to use the methods described in steps 2 and 3.
- the multi-dimensional feature vector analysis technique based on the electromagnetic field simulation database of the dark-field imaging electromagnetic field of the defect in step 2 obtains the defect regression width w regrs , as follows:
- the simulation data uses the physical size x as the abscissa, and the experimental data uses the pixel p as the abscissa. Both use equations (1) and (2) to take the physical size x as the abscissa.
- the pixel size of the CCD is sp
- the magnification of the imaging system is M
- step 2-6 Take the multi-dimensional feature vector of the experimental data of the normalized gray distribution of the defect to be tested in step 2-4 as the test set, input the regression model obtained in step 2-5 to estimate the width of the defect, and obtain the defect The regression width w regrs .
- the simulation data includes defect shape, defect width, defect scattered light intensity distribution I sct (x), and finally constitutes a defect scattered dark field imaging electromagnetic field simulation database.
- step 2-2 the dark-field image of the defect to be detected is collected to obtain the gray distribution along the width direction, and then through deconvolution, the experimental data of the gray distribution under the ideal imaging system after deconvolution is obtained, as follows :
- step 2-2-2-3 Using the PSF measured in step 2-2-2 as the initial value, deconvolve the gray distribution f exp (p) obtained in step 2-2-2 to obtain the ideal image after deconvolution The gray distribution f res (p) under the system.
- the peak value at low grayscale is the gray level of the image background, fill the blank area that appears after rotation with the image background grayscale, and obtain the filled image to maintain the image Completeness
- the multi-dimensional feature vector of the normalized scattered light intensity distribution simulation data and the multi-dimensional feature vector of the normalized gray distribution experimental data described in steps 2-4 are composed of the curve Gaussian pyramid feature of the data, and the extraction process is specific as follows:
- f(n) a certain distribution data
- n discrete data points.
- f(n) is placed on the bottom layer J of the pyramid, denoted as f J (n);
- the regression width w regrs is less than or equal to the system diffraction limit ⁇ 0 , the dark field image of the defect will be seriously degraded, and the effect after deconvolution is not good.
- the grayscale distribution within this width is very close after normalization, and the regression model cannot distinguish To trace the source of the breadth of this range, you need to use the method described in step 3.
- step 3 the dark field image is processed according to the adaptive threshold segmentation method to obtain the defect imaging width w img , as follows:
- the invention proposes a method for real-time traceability of defect width based on the divide and conquer idea, which solves the problems that the width is difficult to measure in real time in the defect detection process and the defect width close to the diffraction limit of the system is difficult to detect.
- This method combines the calibration transfer function, the extreme random tree regression model based on the multi-dimensional feature vector of the defect scattering dark field imaging electromagnetic field simulation database, and the adaptive threshold segmentation algorithm to trace the defect width in real time, and the width recognition has excellent mean square deviation In 0.3 ⁇ m.
- Figure 1 is a general flow chart of the method of the present invention
- Figure 2(a) is the calibrated transfer function under 16 times magnification
- Figure 2(b) shows the calibration error curve at 16 times magnification
- Figure 3 is a flow chart of the multi-dimensional feature vector analysis technology based on the electromagnetic field simulation database of the dark-field imaging electromagnetic field of the defect scattering;
- Figure 4 is a schematic diagram of a curved Gaussian pyramid
- Figure 5(a) is a comparison diagram of traceability width and actual width under 9.7 times magnification
- Figure 5(b) is a schematic diagram of the difference between the traceability width and the actual width at a magnification of 9.7 times;
- Figure 6 (a) is a comparison diagram of the traceability width and the actual width obtained when the traceability method is applied to the dark field image of the defect collected by the Olympus microscope dark field imaging system;
- Figure 6(b) is a schematic diagram of the difference between the traceability width and the actual width obtained when the traceability method is applied to the dark field image of the defect collected by the Olympus microscope dark field imaging system;
- Step 1 Estimate the defect width according to the calibration transfer function of the standard defect digital calibration technology. If the defect width is less than 5 ⁇ m, go to step 2;
- Figure 2 shows the calibration transfer function calibrated at 16 times magnification. It can be seen that the calibration transfer function has a good linear law in the width range above 5 ⁇ m, but it shows a nonlinear law in the width range below 5 ⁇ m. The same pixel The width corresponding to multiple actual widths cannot be distinguished.
- Step 2 According to the multi-dimensional feature vector analysis technology based on the electromagnetic field simulation database of the scattered dark field imaging of the defect, the width of the defect is estimated. The specific process is shown in Figure 3. If the defect width is less than the diffraction limit of the system, go to step 3;
- the simulation data uses the physical size x as the abscissa, and the experimental data uses the pixel p as the abscissa. Both use equations (1) and (2) to take the physical size x as the abscissa.
- the pixel size of the CCD is sp
- the magnification of the imaging system is M
- the curve pyramid is a series of curve data collections arranged in a pyramid shape, derived from the same curve but gradually reduced in resolution, which can realize multi-scale representation of the data. The higher the pyramid, the lower the resolution and size of the curve. The sampling points of the upper curve are half of the lower curve.
- the extraction process of the curved Gaussian pyramid is as follows:
- f(n) a certain distribution data
- n discrete data points.
- f(n) is placed on the bottom layer J of the pyramid, denoted as f J (n);
- step 2-6 Take the multi-dimensional feature vector of the experimental data of the normalized gray distribution of the defect to be tested in step 2-4 as the test set, input the regression model obtained in step 2-5 to estimate the width of the defect, and obtain the defect The regression width w regrs .
- the simulation data includes defect shape, defect width, defect scattered light intensity distribution I sct (x), and finally constitutes a defect scattered dark field imaging electromagnetic field simulation database.
- step 2-2 the dark-field image of the defect to be detected is collected to obtain the gray distribution along the width direction, and then through deconvolution, the experimental data of the gray distribution under the ideal imaging system after deconvolution is obtained, as follows :
- step 2-2-2-3 Using the PSF measured in step 2-2-2 as the initial value, deconvolve the gray distribution f exp (p) obtained in step 2-2-2 to obtain the ideal image after deconvolution The gray distribution f res (p) under the system.
- the peak value at low grayscale is the gray level of the image background, fill the blank area that appears after rotation with the image background grayscale, and obtain the filled image to maintain the image Completeness
- the regression width w regrs is less than or equal to the system diffraction limit ⁇ 0 , the dark field image of the defect will be seriously degraded, and the effect after deconvolution is not good.
- the grayscale distribution within this width is very close after normalization, and the regression model cannot distinguish To trace the source of the breadth of this range, you need to use the method described in step 3.
- Step 3 Process the dark field image according to the adaptive threshold segmentation method to obtain the defect imaging width w img , as follows:
- the invention can effectively trace the source of defects close to the diffraction limit of the system.
- Diffraction limit: ⁇ 0 Use this traceability method to detect 6 defects with a theoretical width of 0.5, 1, 2, 3, 4, 5 ⁇ m on 8 standard boards with depths of 0.028, 0.053, 0.128, 0.149, 0.205, 0.257, 0.503, 1.155 ⁇ m.
- Figure 5 shows the results of the real-time traceability method of defect width based on the divide and conquer idea.
- the abscissa is the sample number
- the discrete data points represent the actual width
- the solid line represents the traceability width
- Figure 5(b) the traceability width
- the traceability error of the width is between ⁇ 0.8 ⁇ m.
- the present invention can effectively trace the source of defects under different dark field systems, as long as the relevant system parameters are acquired and replaced.
- the system parameters that need to be obtained and calculated include:
- ⁇ is the wavelength of the light source of the corresponding dark field detection system
- ⁇ is the standard deviation of the original PSF
- M is the magnification of the original system
- M' is the magnification of other dark field systems
- K is a constant.
- FIG. 6 shows the results of the real-time traceability method of defect width based on the divide and conquer idea.
- the abscissa is the sample number
- the discrete data points represent the actual width
- the solid line represents the traceability width
- Figure 6(b) the traceability width
- the traceability error of the width is within ⁇ 0.65 ⁇ m.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Biochemistry (AREA)
- Pathology (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Geometry (AREA)
- General Health & Medical Sciences (AREA)
- Immunology (AREA)
- Signal Processing (AREA)
- Image Processing (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Image Analysis (AREA)
Abstract
一种基于分治思想的疵病宽度实时溯源方法,即分别运用标定传递函数、基于疵病散射暗场成像电磁场仿真数据库的多维特征向量分析技术、以及自适应阈值分割算法对大于及接近系统衍射极限的疵病宽度进行实时溯源,其中多维特征向量分析技术基于疵病散热暗场成像电磁场仿真数据库中的多维特征向量训练极端随机树回归模型,多维特征向量由数据的曲线高斯金字塔特征构成。解决了疵病检测中疵病宽度难以准确实时测量,以及常规图像处理算法难以准确识别接近系统衍射极限的疵病宽度的问题。易于操作,具有较高的检测效率,宽度溯源的均方偏差优于0.3μm,而且可移植性优异,只需知道系统参数即可方便移植至其它系统中。
Description
本发明属于机器视觉检测技术领域,具体涉及一种基于分治思想的疵病宽度实时溯源方法。
随着工业自动化制造和国际上高精尖技术研究的推进,对精密元件表面质量提出了越来越严格的要求。尤其是在高能激光系统领域,光滑玻璃元件表面疵病的存在可能导致元件炸裂、导致系统无法运行。其中,尤其是大口径元件的表面疵病,由于疵病尺寸相对于元件口径较小且分布随机,是表面质量控制的难点,对疵病的一项十分重要的评价指标就是疵病宽度。为克服传统人工目视检测的效率低、主观性强、易产生歧义等问题,结合机器视觉检测技术和模式识别技术,利用散射暗场成像技术获取大口径元件表面的微米量级宽度的疵病数字图像,针对常规图像处理算法难以对疵病宽度进行实时准确测量的问题,运用标定传递函数进行疵病宽度的实时溯源;针对疵病宽度接近系统衍射极限时标定传递函数难以对疵病宽度进行准确溯源的问题,以疵病散射暗场成像电磁场仿真数据库中的多维特征向量作为数据基础,训练极端随机树回归模型对接近衍射极限的疵病宽度进行溯源;对于宽度更小的疵病则使用自适应阈值分割算法进行溯源。结合以上三种情况,提出一种具有优异可移植性的基于分治思想的疵病宽度实时溯源方法来对疵病宽度进行客观检测和评价。
发明内容
本发明的目的是克服现有技术的不足,提供一种应用于显微散射暗场成像检测环境下的疵病宽度实时溯源方法,提高疵病宽度的识别精度。
本发明解决其技术问题所采用的技术方案包括如下步骤:
步骤1.根据标准疵病数字化标定技术的标定传递函数来估计疵病宽度,记CTF宽度为w
CTF。若w
CTF大于5μm,则此时疵病宽度符合几何成像规律,位于标定传递函数的线性段,使用w
CTF作为疵病实际宽度的估计值较准确;若w
CTF小于等于5μm,则此时疵病宽度较小,成像时会出现衍射增宽现象,不再符合几何成像规律,使用w
CTF作为疵病实际宽度的估计值误差较大,此时跳转到步骤2。
步骤2.根据基于疵病散射暗场成像电磁场仿真数据库的多维特征向量分析技术,估计疵病宽度,记为回归宽度w
regrs。若回归宽度w
regrs大于系统衍射极限σ
0,则使用疵病回归宽度w
regrs作为疵病实际宽度的估计值较准确;若回归宽度w
regrs小于等于系统衍射极限σ
0,则疵病回归宽度w
regrs不能分辨这一宽度范围的不同宽度值,需要使用步骤3进一步识别。
步骤3.根据自适应阈值分割方法处理暗场图像,提取疵病宽度,记为成像宽度w
img。若成像宽度w
img大于系统衍射极限σ
0,则视为奇异点,仍使用回归宽度w
regrs作为实际宽度的估计值;若成像宽度w
img小于等于系统衍射极限σ
0,则使用疵病成像宽度w
img作为实际宽度的估计值。
步骤1所述的标准疵病数字化标定技术,通过建立标定传递函数得到疵病宽度w
CTF。标定传递函数描述疵病的实际宽度与疵病成像后的像素宽度之间的关系,获取过程具体如下:
1-1.根据二元光学原理,在熔石英玻璃基板上,采用电子束曝光与离子束刻蚀技术,刻蚀标准疵病,得到标准板;
1-2.使用扫描电镜(SEM)测量标准板上标准疵病的实际宽度;
1-3.使用暗场成像系统,采集标准板上疵病的暗场灰度图像;
1-4.对暗场灰度图像进行阈值分割、形态学操作等一系列处理,得到疵病的像素宽度;
1-5.以疵病暗场图像的像素宽度为自变量,以疵病SEM测量的实际宽度为因变量,使用最小二乘法对自变量和因变量拟合,得到物像关联函数即为标定传递函数。
由几何光学的物像关系可知,疵病像素宽度与实际宽度呈线性关系,因此通过标定传递函数可以对绝大多数疵病实现宽度溯源。但是随着疵病宽度变小,逼近系统衍射极限,此时几何光学成像理论不再适用,微细疵病将会引起衍射效应,像素宽度会出现增宽现象。这将导致利用标定传递函数进行宽度溯源时,同一像素宽度对应不止一个实际宽度,从而造成混淆而无法准确得到实际宽度。实验发现,疵病宽度小于等于5μm时,标定传递函数不再满足线性关系,宽度溯源出现混淆,对这一范围的宽度溯源需要使用步骤2与步骤3所述方法。
步骤2所述基于疵病散射暗场成像电磁场仿真数据库的多维特征向量分析 技术,得到疵病回归宽度w
regrs,具体如下:
2-1.建立疵病散射暗场成像电磁场仿真数据库,得到多个疵病的散射光强分布的仿真数据;
2-2.使用暗场成像系统采集待测疵病的暗场图像,得到沿宽度方向的灰度分布,然后经过反卷积,得到反卷积后理想成像系统下的灰度分布的实验数据;
2-3.对步骤2-1得到的仿真数据与步骤2-2得到的实验数据,进行坐标系统一与Min-Max归一化,得到归一化后的散射光强分布仿真数据与归一化后的灰度分布实验数据,具体如下:
2-3-1.仿真数据以物理尺寸x为横坐标,实验数据以像素p为横坐标,两者通过式(1)和式(2)将坐标系统一为以物理尺寸x为横坐标,
其中CCD的像元大小为s
p,成像系统的放大倍数为M;
2-3-2.对仿真数据与实验数据进行Min-Max归一化,如式(3)所示,
2-4.分别提取归一化后的散射光强分布仿真数据的多维特征向量与归一化后的灰度分布实验数据的多维特征向量;
2-5.以步骤2-4中归一化后仿真数据的多维特征向量为样本集,采用机器学习理论中有监督学习的极端随机树回归算法,训练得到疵病宽度的回归模型;
2-6.以步骤2-4中待测疵病归一化后灰度分布的实验数据的多维特征向量为测试集,输入步骤2-5中得到的回归模型估计疵病宽度,得到疵病的回归宽度w
regrs。
步骤2-1所述的建立疵病散射暗场成像电磁场仿真数据库,得到多个疵病的散射光强分布的仿真数据,具体如下:
2-1-1.利用FDTD仿真软件建立疵病显微散射暗场成像的电磁场仿真模型;
2-1-2.建立常用疵病的仿真模型;
2-1-3.利用FDTD仿真软件进行电磁场数值计算,得到疵病散射光强分布;
2-1-4.进行近场-远场投射,获得疵病在理想光学成像系统远场像面上光强分布,记为I
sct(x);
2-1-5.改变疵病的形状、尺寸,不断重复步骤2-1-3至2-1-4,获得不同疵病的形状、尺寸对应的疵病散射光强分布,从而获得一系列仿真数据,所述的仿真数据包括疵病形状、疵病宽度、疵病散射光强分布I
sct(x),最终构成疵病散射暗场成像电磁场仿真数据库。
步骤2-2所述的采集待测疵病暗场图像,得到沿宽度方向的灰度分布,然后经过反卷积,得到反卷积后理想成像系统下的灰度分布的实验数据,具体如下:
2-2-1.使用暗场成像系统采集待测疵病的暗场图像,提取疵病暗场图像的沿宽度方向的灰度分布f
exp(p);
2-2-2.使用棋盘格图测量暗场成像系统的点扩散函数(PSF),由于实际光学系统的PSF趋于高斯函数形式,所以在测量过程中使用高斯函数拟合来得到PSF的表达式;
2-2-3.以步骤2-2-2测量得到的PSF为初始值,对步骤2-2-2得到的灰度分布f
exp(p)进行反卷积,得到反卷积后理想成像系统下的灰度分布f
res(p)。
步骤2-2-1所述的提取疵病暗场图像的沿宽度方向的灰度分布,具体如下:
1)对原始暗场图像进行Otsu阈值分割,选取长条形结构元进行开操作连接断裂、孔洞的线特征同时去除噪点,之后提取长直线特征即为疵病特征;
2)使用最小二乘法,对组成该线性特征的众多像素点的像素坐标(u
i,v
i)进行线性拟合,得到直线表达式av+b=u;其中,a、b均为参数;
3)通过参数a可得疵病与v轴正向夹角γ=arctan(a),将原始暗场图像绕左上角旋转-γ即可将疵病旋转为竖直方向;
4)计算原始暗场图像的灰度直方图,低灰度下的峰值为图像背景的灰度水平,用图像背景灰度填充旋转后出现的空白区域,获得填充后的图像,以保持图像的完整性;
5)取填充后图像的列平均值为疵病沿宽度方向的灰度分布f
0(p),呈现单峰型或双峰型。取单峰型分布的峰值横坐标p
0所在的竖直直线为对称轴;取双峰型分布的双峰之间凹陷处的极小值点p
0所在的竖直直线为对称轴;假定数据宽度为2N+1,截取p
0-N~p
0+N之间的数据点为灰度分布f
N(p);
6)将灰度分布f
N(p)轴对称后平均,变为轴对称分布f
sym(p),即
并将坐标轴平移至以对称轴横坐标为坐标原点,得到疵病暗场图像沿宽度方向的灰度分布f
exp(p),即,
f
exp(p)=f
sym(p-p
0),p∈[-N,N] (5)
步骤2-4所述的归一化后散射光强分布仿真数据的多维特征向量和归一化后灰度分布实验数据的多维特征向量,均由数据的曲线高斯金字塔特征构成,其提取过程具体如下:
2-4-1.记某一分布数据为f(n),n表示离散的数据点。f(n)放在金字塔的底层J层,记为f
J(n);
2-4-2.记金字塔第j层曲线分布为f
j(n),对f
j(n)进行低通高斯滤波,得到f′
j(n);
2-4-3.对低通高斯滤波后的曲线分布f′
j(n)以2为因数下采样得到金字塔第j-1层,可以看成是每间隔1个数据点丢弃1个数据点,即
f
j-1(n)=f
2(n)=f
j(2n) (6)
2-4-4.迭代步骤2-4-2、2-4-3 P次,其中j=J,J-1,…,J-P+1,得到P+1层高斯金字塔;
2-4-5.将P+1层高斯金字塔数据串联起来,即为数据的多维特征向量。
若回归宽度w
regrs小于等于系统衍射极限σ
0,此时疵病暗场图像退化严重,反卷积后效果不佳,该宽度范围内的灰度分布归一化后十分接近,回归模型无法分辨,对这一范围的宽度溯源需要使用步骤3所述方法。
步骤3所述的根据自适应阈值分割方法处理暗场图像,得到疵病成像宽度w
img,具体如下:
3-1.使用Otsu算法对疵病暗场图像进行阈值分割;
3-2.提取疵病特征的面积A与长度l,得到疵病的像素宽度w
p=A/l;
3-3.利用CCD的像元大小s
p,成像系统的放大倍数M,得到简单成像宽度w′
img,
3-4.线性拟合位于衍射极限σ
0以下的疵病简单成像宽度w′
img与实际宽度,得到二者之间的关联函数,根据此函数可以修正疵病成像宽度的偏差,据此得到成像宽度w
img。
本发明有益效果如下:
本发明提出的一种基于分治思想的疵病宽度实时溯源方法,解决了疵病检测过程中宽度难以实时测量以及接近系统衍射极限的疵病宽度难以检测的问题。该方法综合了标定传递函数、基于疵病散射暗场成像电磁场仿真数据库的多维特征向量的极端随机树回归模型、自适应阈值分割算法来对疵病宽度进行实时溯源,宽度识别的均方偏差优于0.3μm。该方法不仅可以迁移到其它系统,所构建的疵病散射暗场成像电磁场仿真数据库,还可以方便地与今后的物联网、大数据分析相结合,有利于推动疵病检测技术向智能化、服务化方向发展。
图1为本发明方法的总流程图;
图2(a)为16倍放大倍率下的标定传递函数;
图2(b)为16倍放大倍率下的标定误差曲线;
图3为基于疵病散射暗场成像电磁场仿真数据库的多维特征向量分析技术流程图;
图4为曲线高斯金字塔示意图;
图5(a)为9.7倍放大倍率下溯源宽度与实际宽度对比图;
图5(b)为9.7倍放大倍率下溯源宽度与实际宽度的差值示意图;
图6(a)为将本溯源方法应用于Olympus显微镜暗场成像系统采集的疵病暗场图像时所得溯源宽度与实际宽度对比图;
图6(b)为将本溯源方法应用于Olympus显微镜暗场成像系统采集的疵病暗场图像时所得溯源宽度相对实际宽度的差值示意图;
下面结合附图与具体实施例对本发明作进一步详细描述。
参见附图1,一种基于分治思想的疵病宽度实时溯源方法,具体包括如下步 骤:
步骤1、根据标准疵病数字化标定技术的标定传递函数,估计疵病宽度,若疵病宽度小于5μm,则执行步骤2;
1-1.根据二元光学原理,在熔石英玻璃基板上,采用电子束曝光与离子束刻蚀技术,刻蚀标准疵病,得到标准板;
1-2.使用扫描电镜(SEM)测量标准板上标准疵病的实际宽度;
1-3.使用暗场成像系统,采集标准板上疵病的暗场灰度图像;
1-4.对暗场灰度图像进行阈值分割、形态学操作等一系列处理,得到疵病的像素宽度;
1-5.以疵病暗场图像的像素宽度为自变量,以疵病SEM测量的实际宽度为因变量,使用最小二乘法对自变量和因变量拟合,得到物像关联函数即为标定传递函数。
如图2所示为16倍放大倍率下标定的标定传递函数,能够看出标定传递函数在5μm以上的宽度范围具有很好的线性规律,但在5μm以下的宽度范围呈现非线性规律,同一像素宽度对应多个实际宽度无法分辨。
步骤2、根据基于疵病散射暗场成像电磁场仿真数据库的多维特征向量分析技术,估计疵病宽度,具体流程如图3所示。若疵病宽度小于系统的衍射极限,则执行步骤3;
2-1.建立疵病散射暗场成像电磁场仿真数据库,得到多个疵病的散射光强分布的仿真数据;
2-2.使用暗场成像系统采集待测疵病的暗场图像,得到沿宽度方向的灰度分布,然后经过反卷积,得到反卷积后理想成像系统下的灰度分布的实验数据;
2-3.对步骤2-1得到的仿真数据与步骤2-2得到的实验数据,进行坐标系统一与Min-Max归一化,得到归一化后的散射光强分布仿真数据与归一化后的灰度分布实验数据,具体如下:
2-3-1.仿真数据以物理尺寸x为横坐标,实验数据以像素p为横坐标,两者通过式(1)和式(2)将坐标系统一为以物理尺寸x为横坐标,
其中CCD的像元大小为s
p,成像系统的放大倍数为M;
2-3-2.对仿真数据与实验数据进行Min-Max归一化,如式(3)所示,
2-4.分别提取归一化后的散射光强分布仿真数据与归一化后的灰度分布实验数据的多维特征向量,两者均由数据的曲线高斯金字塔特征构成,如图4所示,曲线金字塔是一系列以金字塔形状排列,来源于同一曲线但分辨率逐渐降低的曲线数据集合,可以实现对数据的多尺度表征。金字塔越往上层,曲线的分辨率和尺寸逐层降低,上层曲线的采样点数是下层曲线的一半。曲线高斯金字塔的提取过程具体如下:
2-4-1.记某一分布数据为f(n),n表示离散的数据点。f(n)放在金字塔的底层J层,记为f
J(n);
2-4-2.记金字塔第j层曲线分布为f
j(n),对f
j(n)进行低通高斯滤波,得到f′
j(n);
2-4-3.对高斯滤波后的曲线分布f′
j(n)以2为因数下采样得到金字塔第j-1层,可以看成是每间隔1个数据点丢弃1个数据点,即
f
j-1(n)=f
2(n)=f
j(2n) (4)
2-4-4.迭代步骤2-4-2、2-4-3P次,其中j=J,J-1,…,J-P+1,得到P+1层高斯金字塔;
2-4-5.将P+1层高斯金字塔数据串联起来,即为数据的多维特征向量。
2-5.以步骤2-4中归一化后仿真数据的多维特征向量为样本集,采用机器学习理论中有监督学习的极端随机树回归算法,训练得到疵病宽度的回归模型;
2-6.以步骤2-4中待测疵病归一化后灰度分布的实验数据的多维特征向量为测试集,输入步骤2-5中得到的回归模型估计疵病宽度,得到疵病的回归宽度w
regrs。
步骤2-1所述的建立疵病散射暗场成像电磁场仿真数据库,得到多个疵病的散射光强分布的仿真数据,具体如下:
2-1-1.利用FDTD仿真软件建立疵病显微散射暗场成像的电磁场仿真模型;
2-1-2.建立常用疵病的仿真模型;
2-1-3.利用FDTD仿真软件进行电磁场数值计算,得到疵病散射光强分布;
2-1-4.进行近场-远场投射,获得疵病在理想光学成像系统远场像面上光强分布,记为I
sct(x);
2-1-5.改变疵病的形状、尺寸,不断重复步骤2-1-3至2-1-4,获得不同疵病的形状、尺寸对应的疵病散射光强分布,从而获得一系列仿真数据,所述的仿真数据包括疵病形状、疵病宽度、疵病散射光强分布I
sct(x),最终构成疵病散射暗场成像电磁场仿真数据库。
步骤2-2所述的采集待测疵病暗场图像,得到沿宽度方向的灰度分布,然后经过反卷积,得到反卷积后理想成像系统下的灰度分布的实验数据,具体如下:
2-2-1.使用暗场成像系统采集待测疵病的暗场图像,提取疵病暗场图像的沿宽度方向的灰度分布f
exp(p);
2-2-2.使用棋盘格图测量暗场成像系统的点扩散函数(PSF),由于实际光学系统的PSF趋于高斯函数形式,所以在测量过程中使用高斯函数拟合来得到PSF的表达式;
2-2-3.以步骤2-2-2测量得到的PSF为初始值,对步骤2-2-2得到的灰度分布f
exp(p)进行反卷积,得到反卷积后理想成像系统下的灰度分布f
res(p)。
步骤2-2-1所述的提取疵病暗场图像的沿宽度方向的灰度分布,具体如下:
1)对原始暗场图像进行Otsu阈值分割,选取长条形结构元进行开操作连接断裂、孔洞的线特征同时去除噪点,之后提取长直线特征即为疵病特征;
2)使用最小二乘法,对组成该线性特征的众多像素点的像素坐标(u
i,v
i)进行线性拟合,得到直线表达式av+b=u;其中,a、b均为参数;
3)通过参数a可得疵病与v轴正向夹角γ=arctan(a),将原始暗场图像绕左上角旋转-γ即可将疵病旋转为竖直方向;
4)计算原始暗场图像的灰度直方图,低灰度下的峰值为图像背景的灰度水平,用图像背景灰度填充旋转后出现的空白区域,获得填充后的图像,以保持图像的完整性;
5)取填充后图像的列平均值为疵病沿宽度方向的灰度分布f
0(p),呈现单峰 型或双峰型。取单峰型分布的峰值横坐标p
0所在的竖直直线为对称轴;取双峰型分布的双峰之间凹陷处的极小值点p
0所在的竖直直线为对称轴;假定数据宽度为2N+1,截取p
0-N~p
0+N之间的数据点为灰度分布f
N(p);
6)将灰度分布f
N(p)轴对称后平均,变为轴对称分布f
sym(p),即
并将坐标轴平移至以对称轴横坐标为坐标原点,得到疵病暗场图像沿宽度方向的灰度分布f
exp(p),即,
f
exp(p)=f
sym(p-p
0),p∈[-N,N] (6)
若回归宽度w
regrs小于等于系统衍射极限σ
0,此时疵病暗场图像退化严重,反卷积后效果不佳,该宽度范围内的灰度分布归一化后十分接近,回归模型无法分辨,对这一范围的宽度溯源需要使用步骤3所述方法。
步骤3、根据自适应阈值分割方法处理暗场图像,得到疵病成像宽度w
img,具体如下:
3-1.使用Otsu算法对疵病暗场图像进行阈值分割;
3-2.提取疵病特征的面积A与长度l,得到疵病的像素宽度w
p=A/l;
3-3.利用CCD的像元大小s
p,成像系统的放大倍数M,得到简单成像宽度w′
img,
3-4.线性拟合位于衍射极限σ
0以下的疵病简单成像宽度w′
img与疵病实际宽度,得到二者之间的关联函数,根据此函数可以修正疵病成像宽度的偏差,据此得到成像宽度w
img。
实施例1:
本发明能够对接近系统衍射极限的疵病进行有效溯源。结合回归宽度w
regrs和成像宽度w
img,以系统的衍射极限为界,暗场成像系统采用白光照明,中心波长λ=550nm,高倍(9.7倍)检测时NA=0.15,可得此时系统的衍射极限:σ
0=
使用本溯源方法对深度分别为0.028、0.053、0.128、0.149、0.205、0.257、0.503、1.155μm的8块标准板上理论宽度分别为0.5、1、2、3、4、5μm的6条疵病进行溯源,可以得到6×8=48条疵病宽度的溯源结果。图5为基于分 治思想的疵病宽度实时溯源方法的结果,图5(a)中横坐标为样本编号,离散数据点表示实际宽度,实线表示溯源宽度,图5(b)表示溯源宽度对实际宽度的估计误差,宽度的溯源误差在±0.8μm之间。
实施例2:
本发明能够对不同暗场系统下的疵病进行有效溯源,只要获取相关系统参数进行替换即可。需要获取和计算的系统参数包括:
1)获取放大倍率和CCD像元大小。在所有需要成像坐标系与像素坐标系转换的地方都会用到系统放大倍率与CCD像元大小,在转换过程中都需要替换;
2)获取数值孔径并计算衍射极限。不同检测系统,具有不同的数值孔径NA,计算不同的衍射极限σ
0,在宽度溯源时加以替换:
其中λ为相应暗场检测系统光源的波长;
3)计算点扩散函数。不同检测系统的PSF不同,重新用棋盘格图标定,不符合快速检测需求。由于反卷积采用的算法不必以先验PSF知识为基础,仅需提供PSF的初始估计,其它暗场系统的PSF计算方法如式(9)所示:
1)放大倍率:
M
Olym=10×0.63=6.3 (10)
2)衍射极限:
3)点扩散函数:
采集深度为0.128、0.149、0.205、0.257、0.503μm标准板上宽度为0.5、1、2、3、4、5μm的疵病的暗场图像,使用经过参数替换后的模型对其进行宽度识别。图6为基于分治思想的疵病宽度实时溯源方法的结果,图6(a)中横坐标为样本编号,离散数据点表示实际宽度,实线表示溯源宽度,图6(b)表示溯源宽度对实际宽度的估计误差,宽度的溯源误差在±0.65μm之间。
Claims (7)
- 一种基于分治思想的疵病宽度实时溯源方法,其特征在于包括如下步骤:步骤1.根据标准疵病数字化标定技术的标定传递函数来估计疵病宽度,记估计的疵病宽度为w CTF;若w CTF大于5μm,则此时疵病宽度符合几何成像规律,位于标定传递函数的线性段,使用w CTF作为疵病实际宽度的估计值较准确;若w CTF小于等于5μm,则此时疵病宽度较小,成像时会出现衍射增宽现象,不再符合几何成像规律,使用w CTF作为疵病实际宽度的估计值误差较大,此时跳转到步骤2;步骤2.根据基于疵病散射暗场成像电磁场仿真数据库的多维特征向量分析技术,估计疵病宽度,记为回归宽度w regrs;若回归宽度w regrs大于系统衍射极限σ 0,则使用疵病回归宽度w regrs作为疵病实际宽度的估计值较准确;若回归宽度w regrs小于等于系统衍射极限σ 0,则疵病回归宽度w regrs不能分辨这一宽度范围的不同宽度值,需要使用步骤3进一步识别;步骤3.根据自适应阈值分割方法处理暗场图像,提取疵病宽度,记为成像宽度w img;若成像宽度w img大于系统衍射极限σ 0,则视为奇异点,仍使用回归宽度w regrs作为实际宽度的估计值;若成像宽度w img小于等于系统衍射极限σ 0,则使用疵病成像宽度w img作为实际宽度的估计值。
- 根据权利要求1所述的一种基于分治思想的疵病宽度实时溯源方法,其特征在于步骤2所述的基于疵病散射暗场成像电磁场仿真数据库的多维特征向量分析技术,具体如下:2-1.建立疵病散射暗场成像电磁场仿真数据库,得到远场CCD像面上多个疵病的散射光强分布的电磁场仿真数据;2-2.采集待测疵病暗场图像,得到沿宽度方向的灰度分布,然后经过反卷积,得到理想成像系统下的灰度分布的实验数据;2-3.对步骤2-1得到的仿真数据与步骤2-2得到的实验数据,进行坐标系统一与Min-Max归一化,得到归一化后的散射光强分布仿真数据与归一化后的灰度分布实验数据;2-4.分别提取归一化后的散射光强分布仿真数据的多维特征向量与归一化后的灰度分布实验数据的多维特征向量;2-5.以步骤2-4中归一化后仿真数据的多维特征向量为样本集,采用机器学 习理论中有监督学习的极端随机树回归算法,训练得到疵病宽度的回归模型;2-6.以步骤2-4中待测疵病归一化后灰度分布的实验数据的多维特征向量为测试集,输入步骤2-5中得到的回归模型所溯源出的疵病宽度,得到疵病的回归宽度w regrs。
- 根据权利要求2所述的基于疵病散射暗场成像电磁场仿真数据库的多维特征向量分析技术,其特征在于步骤2-2所述的反卷积后理想成像系统下灰度分布的实验数据获得过程,具体如下:2-2-1.使用暗场成像系统采集待测疵病的暗场图像,提取疵病暗场图像的沿宽度方向的灰度分布f exp(p);2-2-2.使用棋盘格图来初步测量暗场成像系统的点扩散函数(PSF),由于实际光学系统的PSF趋于高斯函数形式,所以在测量过程中使用高斯函数拟合来得到PSF的表达式;2-2-3.以该点扩散函数作为初始值,来对得到的灰度分布进行反卷积,即可得到反卷积后理想成像系统下的灰度分布f res(p)。
- 根据权利要求3所述的反卷积后理想成像系统下灰度分布的检测数据获得过程,其特征在于步骤2-2-1所述的提取疵病暗场图像的沿宽度方向的灰度分布,具体如下:1)对原始暗场图像进行Otsu阈值分割,选取长条形结构元进行开操作连接断裂、孔洞的线特征同时去除噪点,之后提取长直线特征即为疵病特征;2)使用最小二乘法,对组成该线性特征的众多像素点的像素坐标(u i,v i)进行线性拟合,得到直线表达式av+b=u;其中,a、b均为参数;3)通过参数a可得刻线与v轴正向夹角γ=arctan(a),将原始暗场图像绕左上角旋转-γ即可将疵病旋转为竖直方向;4)计算原始暗场图像的灰度直方图,低灰度下的峰值为图像背景的灰度水平,用图像背景灰度填充旋转后出现的空白区域,获得填充后的图像,以保持图像的完整性;5)取填充后图像的列平均值为疵病沿宽度方向的灰度分布f 0(p),呈现单峰型或双峰型;取单峰型分布的峰值横坐标p 0所在的竖直直线为对称轴;取双峰型分布的双峰之间凹陷处的极小值点p 0所在的竖直直线为对称轴;假定数据宽 度为2N+1,截取p 0-N~p 0+N之间的数据点为灰度分布f N(p);6)将灰度分布f N(p)轴对称后平均,变为轴对称分布f sym(p),即并将坐标轴平移至以对称轴横坐标为坐标原点,得到疵病暗场图像沿宽度方向的灰度分布f exp(p),即,f exp(p)=f sym(p-p 0),p∈[-N,N] (2)。
- 根据权利要求2或4所述的基于疵病散射暗场成像电磁场仿真数据库的多维特征向量分析技术,其特征在于步骤2-4所述的归一化后散射光强分布仿真数据的多维特征向量和归一化后灰度分布实验数据的多维特征向量均由数据的曲线高斯金字塔特征构成,其提取过程具体如下:1)记某一分布数据为f(n),n表示离散的数据点;f(n)放在金字塔的底层J层,记为f J(n);2)记金字塔第j层曲线分布为f j(n),对f j(n)进行低通高斯滤波,得到f′ j(n);3)对高斯滤波后的曲线分布f′ j(n)以2为因数下采样得到金字塔第j-1层,可以看成是每间隔1个数据点丢弃1个数据点,即:f j-1(n)=f 2↓(n)=f j(2n) (3)4)迭代步骤2)、3)P次,其中j=J,J-1,…,J-P+1,得到P+1层高斯金字塔;5)将P+1层高斯金字塔数据串联起来,即为数据的多维特征向量。
- 根据权利要求6所述的一种基于分治思想的疵病宽度实时溯源方法,其特征在于本发明对于其它暗场成像系统下的疵病检测具有可移植性,只需要获取其它暗场成像系统的相关系统参数进行替换即可,具体如下:获取放大倍率与CCD像元大小;在所有需要成像坐标系与像素坐标系转换的地方都会用到系统放大倍率与CCD像元大小,在转换过程中都需要替换;获取数值孔径并计算衍射极限;不同检测系统,具有不同的数值孔径NA,计算不同的衍射极限σ 0,在使用决策模型库时加以替换;计算点扩散函数;不同检测系统的PSF不同,重新用棋盘格图标定,不符合快速检测需求;由于反卷积不必以先验PSF知识为基础,仅需提供PSF的初始估计,其它暗场系统的PSF计算方法如式(5)所示:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US17/266,091 US11636584B2 (en) | 2019-06-14 | 2019-08-02 | Real-time traceability method of width of defect based on divide-and-conquer |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910516180.1 | 2019-06-14 | ||
| CN201910516180 | 2019-06-14 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2020248345A1 true WO2020248345A1 (zh) | 2020-12-17 |
Family
ID=73781596
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2019/099008 Ceased WO2020248345A1 (zh) | 2019-06-14 | 2019-08-02 | 一种基于分治思想的疵病宽度实时溯源方法 |
Country Status (2)
| Country | Link |
|---|---|
| US (1) | US11636584B2 (zh) |
| WO (1) | WO2020248345A1 (zh) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115115640A (zh) * | 2022-08-30 | 2022-09-27 | 南通美迪森医药科技有限公司 | 基于光学手段的胶囊外壳表面缺陷检测方法 |
| CN116563280A (zh) * | 2023-07-07 | 2023-08-08 | 深圳市鑫典金光电科技有限公司 | 一种基于数据分析的复合铜散热底板加工检测方法及系统 |
| CN117392126A (zh) * | 2023-12-08 | 2024-01-12 | 四川省水利科学研究院 | 一种水工隧洞缺陷检测方法 |
| CN120318181A (zh) * | 2025-03-31 | 2025-07-15 | 深圳中科飞测科技股份有限公司 | 一种晶圆表面污染缺陷检出方法及装置 |
| CN121805267A (zh) * | 2026-03-10 | 2026-04-07 | 西安获德图像技术有限公司 | 基于机器视觉的涤纶长丝纺丝在线检测系统及方法 |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114120318B (zh) * | 2021-11-29 | 2024-04-26 | 哈尔滨工业大学 | 一种基于集成决策树的暗场图像目标点精确提取方法 |
| CN114878583B (zh) * | 2022-07-08 | 2022-09-20 | 四川大学 | 用于畸变光斑照明疵病暗场成像的图像处理方法及系统 |
| CN114994060B (zh) * | 2022-07-15 | 2022-11-15 | 君享科技(深圳)有限公司 | 一种机器视觉下的磁环外观缺陷智能检测系统及方法 |
| CN116337879B (zh) * | 2023-05-23 | 2023-08-04 | 青岛豪迈电缆集团有限公司 | 一种电缆绝缘表皮磨损缺陷快速检测方法 |
| CN118886786B (zh) * | 2023-12-11 | 2026-08-11 | 南京安昇超科技有限公司 | 冶金制品质检管理系统 |
| CN118037664B (zh) * | 2024-02-20 | 2024-10-01 | 成都天兴山田车用部品有限公司 | 一种深孔表面缺陷检测与cv尺寸计算方法 |
| CN117934466B (zh) * | 2024-03-21 | 2024-06-04 | 上海渺知科技有限公司 | 一种三维模型辅助电磁光融合成像金属结构检测方法 |
| CN120629174B (zh) * | 2025-07-10 | 2026-03-31 | 扬州鑫润新材料有限公司 | 一种薄膜表面缺陷检测方法及系统 |
| CN122048929B (zh) * | 2026-04-14 | 2026-06-16 | 上海帆声图像科技有限公司 | 一种基于明暗视场线扫相机的分区自适应检测与多尺度缺陷提取方法 |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1563957A (zh) * | 2004-04-09 | 2005-01-12 | 浙江大学 | 光滑表面疵病的自动化检测方法及其系统 |
| CN2729708Y (zh) * | 2004-06-22 | 2005-09-28 | 浙江大学 | 光滑表面疵病的自动化检测系统 |
| US7423746B2 (en) * | 2006-02-28 | 2008-09-09 | Hitachi High-Technologies Corporation | Circuit-pattern inspecting apparatus and method |
| CN102636496A (zh) * | 2012-04-24 | 2012-08-15 | 浙江大学 | 光学表面疵病暗场检测中疵病宽度标定标准化系统及方法 |
| CN104833679A (zh) * | 2015-04-29 | 2015-08-12 | 浙江大学 | 一种微观缺陷三维尺度逆向标定及检测方法 |
| CN105021417A (zh) * | 2015-07-09 | 2015-11-04 | 天津大学 | 数字x射线成像系统调制传递函数的狭缝法测量方法 |
| CN108061529A (zh) * | 2018-02-23 | 2018-05-22 | 西南科技大学 | 基于干涉图像自相关值曲率特征的表面粗糙度测量方法 |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9547892B2 (en) * | 2014-05-06 | 2017-01-17 | Kla-Tencor Corporation | Apparatus and methods for predicting wafer-level defect printability |
| US10599951B2 (en) * | 2018-03-28 | 2020-03-24 | Kla-Tencor Corp. | Training a neural network for defect detection in low resolution images |
-
2019
- 2019-08-02 US US17/266,091 patent/US11636584B2/en active Active
- 2019-08-02 WO PCT/CN2019/099008 patent/WO2020248345A1/zh not_active Ceased
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1563957A (zh) * | 2004-04-09 | 2005-01-12 | 浙江大学 | 光滑表面疵病的自动化检测方法及其系统 |
| CN2729708Y (zh) * | 2004-06-22 | 2005-09-28 | 浙江大学 | 光滑表面疵病的自动化检测系统 |
| US7423746B2 (en) * | 2006-02-28 | 2008-09-09 | Hitachi High-Technologies Corporation | Circuit-pattern inspecting apparatus and method |
| CN102636496A (zh) * | 2012-04-24 | 2012-08-15 | 浙江大学 | 光学表面疵病暗场检测中疵病宽度标定标准化系统及方法 |
| CN104833679A (zh) * | 2015-04-29 | 2015-08-12 | 浙江大学 | 一种微观缺陷三维尺度逆向标定及检测方法 |
| CN105021417A (zh) * | 2015-07-09 | 2015-11-04 | 天津大学 | 数字x射线成像系统调制传递函数的狭缝法测量方法 |
| CN108061529A (zh) * | 2018-02-23 | 2018-05-22 | 西南科技大学 | 基于干涉图像自相关值曲率特征的表面粗糙度测量方法 |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115115640A (zh) * | 2022-08-30 | 2022-09-27 | 南通美迪森医药科技有限公司 | 基于光学手段的胶囊外壳表面缺陷检测方法 |
| CN116563280A (zh) * | 2023-07-07 | 2023-08-08 | 深圳市鑫典金光电科技有限公司 | 一种基于数据分析的复合铜散热底板加工检测方法及系统 |
| CN116563280B (zh) * | 2023-07-07 | 2023-09-12 | 深圳市鑫典金光电科技有限公司 | 一种基于数据分析的复合铜散热底板加工检测方法及系统 |
| CN117392126A (zh) * | 2023-12-08 | 2024-01-12 | 四川省水利科学研究院 | 一种水工隧洞缺陷检测方法 |
| CN117392126B (zh) * | 2023-12-08 | 2024-03-15 | 四川省水利科学研究院 | 一种水工隧洞缺陷检测方法 |
| CN120318181A (zh) * | 2025-03-31 | 2025-07-15 | 深圳中科飞测科技股份有限公司 | 一种晶圆表面污染缺陷检出方法及装置 |
| CN120318181B (zh) * | 2025-03-31 | 2026-02-24 | 深圳中科飞测科技股份有限公司 | 一种晶圆表面污染缺陷检出方法及装置 |
| CN121805267A (zh) * | 2026-03-10 | 2026-04-07 | 西安获德图像技术有限公司 | 基于机器视觉的涤纶长丝纺丝在线检测系统及方法 |
Also Published As
| Publication number | Publication date |
|---|---|
| US11636584B2 (en) | 2023-04-25 |
| US20210312609A1 (en) | 2021-10-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11636584B2 (en) | Real-time traceability method of width of defect based on divide-and-conquer | |
| CN115482195B (zh) | 一种基于三维点云的列车部件变形检测方法 | |
| CN107677216B (zh) | 一种基于光度立体视觉的多个磨粒三维形貌同步获取方法 | |
| US8311314B2 (en) | Pattern measuring method and pattern measuring device | |
| CN108428231B (zh) | 一种基于随机森林的多参数零件表面粗糙度学习方法 | |
| CN109716495B (zh) | 用于晶片中开口尺寸的光学测量的方法和系统 | |
| CN108257171A (zh) | 基于光视觉的汽车雷达装配孔径检测方法 | |
| JP5913903B2 (ja) | 形状検査方法およびその装置 | |
| WO2013061976A1 (ja) | 形状検査方法およびその装置 | |
| CN117252915A (zh) | 一种基于改进梯度加权的零件图像高精度聚焦方法及装置 | |
| CN113222955A (zh) | 一种基于机器视觉的齿轮尺寸参数自动测量方法 | |
| CN118882520B (zh) | 一种大口径曲面光学元件表面缺陷三维检测装置及方法 | |
| CN119291910A (zh) | 一种显微镜自动对焦方法、装置和计算机可读存储介质 | |
| Yang et al. | Surface defects evaluation system based on electromagnetic model simulation and inverse-recognition calibration method | |
| CN121921583A (zh) | 基于图像增强技术的螺栓表面微小缺陷特征提取方法 | |
| CN106017328A (zh) | 一种多类型线宽测量方法及装置 | |
| Hao et al. | An image-based hairline crack identification method for metal parts | |
| CN112414316B (zh) | 一种应变片敏感栅尺寸参数测量方法 | |
| KR20250132000A (ko) | 반도체 계측을 위한 학습기반 의미적 분할 방법 및 장치 | |
| CN120876579B (zh) | 用于光学元件厚度检测的sd-oct系统及其检测方法 | |
| CN121962076A (zh) | 灰铸铁铸件表面气孔与裂纹缺陷多角度视觉融合检测方法 | |
| CN119495587B (zh) | 一种晶圆贴膜瑕疵检测系统及方法 | |
| Yang et al. | Sparse microdefect evaluation system for large fine optical surfaces based on dark-field microscopic scattering imaging | |
| CN121073891A (zh) | 一种基于计算机视觉的车损检测识别方法 | |
| CN119290891B (zh) | 一种基于SOD-YOLOv5模型的靶丸定位圆测量缺陷检测方法及装置 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19932940 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 19932940 Country of ref document: EP Kind code of ref document: A1 |



